sh2(A) exception handler update

PATCH Applied Yoshinori Sato 2 participants 6 messages Jul 9, 2008 – Jul 28, 2008 SuperH

lore

Share & tools

lore

Yoshinori Sato

author maintainer

2008-07-09 16:20 UTC #0 lore raw

PATCH 1/1 sh2(A) exception handler update

This patch is

By sh2
- Remove duplicate code
- Reduce stack usage
- Cleanup and little optimize

By sh2a
- Add missing handler(256 to 511)
- Use sh2a instructions handler

Signed-off-by Yoshinori Sato <[email protected]>

arch/sh/kernel/cpu/sh2/entry.S +51 ~~-82~~

index 0fc8906..ee894e5 100644
@@ -3,7 +3,7 @@
  *
  * The SH-2 exception entry
  *
- * Copyright (C) 2005,2006 Yoshinori Sato
+ * Copyright (C) 2005-2008 Yoshinori Sato
  * Copyright (C) 2005  AXE,Inc.
  *
  * This file is subject to the terms and conditions of the GNU General Public

@@ -36,43 +36,41 @@ OFF_TRA	=  (16*4+6*4)
 #include <asm/entry-macros.S>
 
 ENTRY(exception_handler)
-	! already saved r0/r1
+	! stack
+	! r0 <- point sp
+	! r1
+	! pc
+	! sr
+	! r0 = temporary
+	! r1 = vector (pseudo EXPEVT / INTEVT / TRA)
 	mov.l	r2,@-sp
 	mov.l	r3,@-sp
-	mov	r0,r1
 	cli
 	mov.l	$cpu_mode,r2
 	mov.l	@r2,r0
 	mov.l	@(5*4,r15),r3	! previous SR
-	shll2	r3		! set "S" flag
-	rotl	r0		! T <- "S" flag
-	rotl	r0		! "S" flag is LSB
-	rotcr	r3		! T -> r3:b30
-	shlr	r3
-	shlr	r0
-	bt/s	1f
-	 mov.l	r3,@(5*4,r15)	! copy cpu mode to SR
+	or	r0,r3		! set MD
+	tst	r0,r0
+	bf/s	1f		! previous mode check
+	 mov.l	r3,@(5*4,r15)	! update SR
 	! switch to kernel mode
-	mov	#1,r0
-	rotr	r0
-	rotr	r0
+	mov.l	__md_bit,r0
 	mov.l	r0,@r2		! enter kernel mode
 	mov.l	$current_thread_info,r2
 	mov.l	@r2,r2
-	mov	#0x20,r0
+	mov	#(THREAD_SIZE >> 8),r0
 	shll8	r0
 	add	r2,r0
 	mov	r15,r2		! r2 = user stack top
 	mov	r0,r15		! switch kernel stack
-	add	#-4,r15		! dummy
 	mov.l	r1,@-r15	! TRA
 	sts.l	macl, @-r15
 	sts.l	mach, @-r15
 	stc.l	gbr, @-r15
-	mov.l	@(4*4,r2),r0
-	mov.l	@(5*4,r2),r1
-	mov.l	r1,@-r15	! original SR
+	mov.l	@(5*4,r2),r0
+	mov.l	r0,@-r15	! original SR
 	sts.l	pr,@-r15
+	mov.l	@(4*4,r2),r0
 	mov.l	r0,@-r15	! original PC
 	mov	r2,r3
 	add	#(4+2)*4,r3	! rewind r0 - r3 + exception frame

@@ -88,14 +86,15 @@ ENTRY(exception_handler)
 	mov.l	r6,@-r15
 	mov.l	r5,@-r15
 	mov.l	r4,@-r15
+	mov	r1,r9		! save TRA
 	mov	r2,r8		! copy user -> kernel stack
-	mov.l	@r8+,r3
+	mov.l	@(0,r8),r3
 	mov.l	r3,@-r15
-	mov.l	@r8+,r2
+	mov.l	@(4,r8),r2
 	mov.l	r2,@-r15
-	mov.l	@r8+,r1
+	mov.l	@(12,r8),r1
 	mov.l	r1,@-r15
-	mov.l	@r8+,r0
+	mov.l	@(8,r8),r0
 	bra	2f
 	 mov.l	r0,@-r15
 1:

@@ -107,10 +106,11 @@ ENTRY(exception_handler)
 	mov.l	r0,@-r15	
 	mov.l	@r2+,r0		! old R2
 	mov.l	r0,@-r15	
-	mov.l	@r2+,r0		! old R1
-	mov.l	r0,@-r15	
-	mov.l	@r2+,r0		! old R0
+	mov.l	@(4,r2),r0	! old R1
 	mov.l	r0,@-r15	
+	mov.l	@r2,r0		! old R0
+	mov.l	r0,@-r15
+	add	#8,r2
 	mov.l	@r2+,r3		! old PC
 	mov.l	@r2+,r0		! old SR
 	add	#-4,r2		! exception frame stub (sr)

@@ -135,14 +135,12 @@ ENTRY(exception_handler)
 	mov.l	r6,@-r2
 	mov.l	r5,@-r2
 	mov.l	r4,@-r2
+	mov	r1,r9
 	mov.l	@(OFF_R0,r15),r0
 	mov.l	@(OFF_R1,r15),r1
 	mov.l	@(OFF_R2,r15),r2
 	mov.l	@(OFF_R3,r15),r3
 2:
-	mov	#OFF_TRA,r8
-	add	r15,r8
-	mov.l	@r8,r9	
 	mov	#64,r8
 	cmp/hs	r8,r9
 	bt	interrupt_entry	! vec >= 64 is interrupt

@@ -150,26 +148,14 @@ ENTRY(exception_handler)
 	cmp/hs	r8,r9
 	bt	trap_entry	! 64 > vec >= 32  is trap
 
-#if defined(CONFIG_SH_FPU)
-	mov     #13,r8
-	cmp/eq  r8,r9
-	bt      10f             ! fpu
-	nop
-#endif
-
 	mov.l	4f,r8
 	mov	r9,r4
 	shll2	r9
 	add	r9,r8
-	mov.l	@r8,r8
-	mov	#0,r9
-	cmp/eq	r9,r8
+	mov.l	@r8,r8		! exception handler address
+	tst	r8,r8
 	bf	3f
 	mov.l	8f,r8		! unhandled exception
-#if defined(CONFIG_SH_FPU)
-10:
-	mov.l	9f, r8		! unhandled exception
-#endif
 3:
 	mov.l	5f,r10
 	jmp	@r8

@@ -188,10 +174,7 @@ interrupt_entry:
 5:	.long	ret_from_exception
 6:	.long	ret_from_irq
 7:	.long	do_IRQ
-8:	.long	do_exception_error
-#ifdef CONFIG_SH_FPU
-9:	.long	fpu_error_trap_handler
-#endif
+8:	.long	exception_error
 
 trap_entry:
 	mov	#0x30,r8

@@ -200,24 +183,9 @@ trap_entry:
 	add	#-0x10,r9	! convert SH2 to SH3/4 ABI
 1:	
 	shll2	r9			! TRA
-	mov	#OFF_TRA,r8
-	add	r15,r8
-	mov.l	r9,@r8
-	mov	r9,r8
-#ifdef CONFIG_TRACE_IRQFLAGS
-	mov.l	2f, r9
-	jsr	@r9
-	 nop
-#endif
-	sti
-	bra	system_call
-	 nop
+	bra	system_call	! jump common systemcall entry
+	 mov	r9,r8
 	
-	.align	2
-#ifdef CONFIG_TRACE_IRQFLAGS
-2:	.long	trace_hardirqs_on
-#endif
-
 #if defined(CONFIG_SH_STANDARD_BIOS)
 	/* Unwind the stack and jmp to the debug entry */
 ENTRY(sh_bios_handler)

@@ -240,7 +208,7 @@ ENTRY(sh_bios_handler)
 	mov.l	@r2,r2
 	stc	sr,r3
 	mov.l	r2,@r0
-	mov.l	r3,@r0
+	mov.l	r3,@(4,r0)
 	mov.l	r1,@(8,r0)	
 	mov.l	@r15+, r0
 	mov.l	@r15+, r1

@@ -272,22 +240,30 @@ ENTRY(address_error_trap_handler)
 	mov.l	1f,r0
 	jmp	@r0
 	 mov	#0,r5				! writeaccess is unknown
-	.align	2
 
+	.align	2
 1:	.long	do_address_error
 
 restore_all:
-	cli
-#ifdef CONFIG_TRACE_IRQFLAGS
-	mov.l	1f, r0
-	jsr	@r0
-	 nop
-#endif
+	stc	sr,r0
+	or	#0xf0,r0
+	ldc	r0,sr				! all interrupt block (same BL = 1)
+	! restore special register
+	! overlap exception frame
+	mov	r15,r0
+	add	#17*4,r0
+	lds.l	@r0+,pr
+	add	#4,r0
+	ldc.l	@r0+,gbr
+	lds.l	@r0+,mach
+	lds.l	@r0+,macl
 	mov	r15,r0
 	mov.l	$cpu_mode,r2
 	mov	#OFF_SR,r3
 	mov.l	@(r0,r3),r1
-	mov.l	r1,@r2
+	mov.l	__md_bit,r3
+	and	r1,r3				! copy MD bit
+	mov.l	r3,@r2
 	shll2	r1				! clear MD bit
 	shlr2	r1
 	mov.l	@(OFF_SP,r0),r2

@@ -297,12 +273,6 @@ restore_all:
 	mov	#OFF_PC,r3
 	mov.l	@(r0,r3),r1
 	mov.l	r1,@r2				! set pc
-	add	#4*16+4,r0
-	lds.l	@r0+,pr
-	add	#4,r0				! skip sr
-	ldc.l	@r0+,gbr
-	lds.l	@r0+,mach
-	lds.l	@r0+,macl
 	get_current_thread_info r0, r1
 	mov.l	$current_thread_info,r1
 	mov.l	r0,@r1

@@ -326,9 +296,8 @@ restore_all:
 	 nop
 
 	.align 2
-#ifdef CONFIG_TRACE_IRQFLAGS
-1:     .long   trace_hardirqs_off
-#endif
+__md_bit:
+	.long	0x40000000
 $current_thread_info:
 	.long	__current_thread_info
 $cpu_mode:

arch/sh/kernel/cpu/sh2/ex.S +8 -7

index 6d285af..85b0bf8 100644
@@ -18,16 +18,17 @@
 exception_entry:	
 no	=	0
 	.rept	256
-	mov.l	r0,@-sp
-	mov	#no,r0
+	mov.l	r1,@-sp
 	bra	exception_trampoline
-	and	#0xff,r0
+	mov	#no,r1
 no	=	no + 1
 	.endr
 exception_trampoline:
-	mov.l	r1,@-sp
-	mov.l	$exception_handler,r1
-	jmp	@r1
+	mov.l	r0,@-sp
+	mov.l	$exception_handler,r0
+	extu.b	r1,r1
+	jmp	@r0
+	  extu.w	r1,r1
 
 	.align	2
 $exception_entry:

@@ -41,6 +42,6 @@ $exception_handler:
 ENTRY(vbr_base)
 vector	=	0
 	.rept	256
-	.long	exception_entry + vector * 8
+	.long	exception_entry + vector * 6
 vector	=	vector + 1
 	.endr

arch/sh/kernel/cpu/sh2a/Makefile +1 -1

index 7e2b90c..1ab1ecf 100644
@@ -4,7 +4,7 @@
 
 obj-y	:= common.o probe.o opcode_helper.o
 
-common-y	+= $(addprefix ../sh2/, ex.o entry.o)
+common-y	+= ex.o entry.o
 
 obj-$(CONFIG_SH_FPU)	+= fpu.o

arch/sh/kernel/cpu/sh2a/entry.S new file +249 -0

249 lines not on this page · raw message

arch/sh/kernel/cpu/sh2a/ex.S new file +72 -0

72 lines not on this page · raw message

arch/sh/kernel/traps_32.c +4 -0

index e08b3bf..511a942 100644
@@ -43,6 +43,7 @@
 # define TRAP_ILLEGAL_SLOT_INST	6
 # define TRAP_ADDRESS_ERROR	9
 # ifdef CONFIG_CPU_SH2A
+#  define TRAP_FPU_ERROR	13
 #  define TRAP_DIVZERO_ERROR	17
 #  define TRAP_DIVOVF_ERROR	18
 # endif

@@ -851,6 +852,9 @@ void __init trap_init(void)
 #ifdef CONFIG_CPU_SH2A
 	set_exception_table_vec(TRAP_DIVZERO_ERROR, do_divide_error);
 	set_exception_table_vec(TRAP_DIVOVF_ERROR, do_divide_error);
+#ifdef CONFIG_SH_FPU
+	set_exception_table_vec(TRAP_FPU_ERROR, fpu_error_trap_handler);
+#endif
 #endif
 
 	/* Setup VBR for boot cpu */

Signature · 2 lines

Yoshinori Sato
<[email protected]>

Paul Mundt

Applied

2008-07-17 05:55 UTC #1 lore raw

Quoted 7 lines · Yoshinori Sato

This patch is
By sh2
- Remove duplicate code
- Reduce stack usage
- Cleanup and little optimize

Quoted 6 lines

By sh2a
- Add missing handler(256 to 511)
- Use sh2a instructions handler
Signed-off-by Yoshinori Sato <[email protected]>

Applied, thanks.

Paul Mundt

2008-07-17 07:09 UTC #2 lore raw

In reply to Yoshinori Sato

Quoted 7 lines · Yoshinori Sato

This patch is
By sh2
- Remove duplicate code
- Reduce stack usage
- Cleanup and little optimize

Quoted 6 lines

By sh2a
- Add missing handler(256 to 511)
- Use sh2a instructions handler
Signed-off-by Yoshinori Sato <[email protected]>

There still seems to be something buggy here, which may or may not be toolchain related. Booting the current git kernel, does the 'ps aux' output look sane on your board?

Yoshinori Sato

author maintainer

2008-07-17 14:46 UTC #3 lore raw

In reply to Yoshinori Sato

At Thu, 17 Jul 2008 16:09:43 +0900,

Quoted 13 lines · Paul Mundt

Quoted 11 lines · Yoshinori Sato
This patch is
By sh2
- Remove duplicate code
- Reduce stack usage
- Cleanup and little optimize
By sh2a
- Add missing handler(256 to 511)
- Use sh2a instructions handler

Quoted 5 lines

Quoted 2 lines
Signed-off-by Yoshinori Sato <[email protected]>
There still seems to be something buggy here, which may or may not be toolchain related. Booting the current git kernel, does the 'ps aux' output look sane on your board?

Hmm...
'ps aux' is unfinished.
proc readdir is return non NULL always.

I check it.

Signature · 2 lines

Yoshinori Sato
<[email protected]>

Yoshinori Sato

author maintainer

2008-07-19 19:35 UTC #4 lore raw

In reply to Yoshinori Sato

At Thu, 17 Jul 2008 10:46:34 -0400,

Quoted 22 lines · Yoshinori Sato

At Thu, 17 Jul 2008 16:09:43 +0900,
Quoted 13 lines · Paul Mundt
Quoted 11 lines · Yoshinori Sato
This patch is
By sh2
- Remove duplicate code
- Reduce stack usage
- Cleanup and little optimize
By sh2a
- Add missing handler(256 to 511)
- Use sh2a instructions handler
Quoted 5 lines
Quoted 2 lines
Signed-off-by Yoshinori Sato <[email protected]>
There still seems to be something buggy here, which may or may not be toolchain related. Booting the current git kernel, does the 'ps aux' output look sane on your board?

Quoted 5 lines

Hmm...
'ps aux' is unfinished.
proc readdir is return non NULL always.
I check it.

OK. I'm fixed.

I think this problem is GCC(4.1.2) bug.
Syscall "getdents" returned "dirent->d_off" is always 0. I think other EB enviroment have same problem.

Problem code
0c03c954 <filldir>:
:

 c03c97a:       58 f7           mov.l   @(28,r15),r8 !-> offset (high)
 c03c97c:       59 f8           mov.l   @(32,r15),r9 !-> offset (low)
 c03c97e:       53 f9           mov.l   @(36,r15),r3
 c03c980:       54 fa           mov.l   @(40,r15),r4

:
c03c9a0: 21 82 mov.l r8,@r1 !offset(high) -> dirent->d_off

It's workaround patch.

Signed-off-by Yoshinori Sato <[email protected]>

include/asm-sh/uaccess_32.h +1 -1

index ae0d24f..d3baea8 100644
@@ -76,7 +76,7 @@ do {							\
 		__put_user_asm(x, ptr, retval, "w");	\
 		break;					\
 	case 4:						\
-		__put_user_asm(x, ptr, retval, "l");	\
+		__put_user_asm((u32)x, ptr, retval, "l");	\
 		break;					\
 	case 8:						\
 		__put_user_u64(x, ptr, retval);		\

Signature · 2 lines

Yoshinori Sato
<[email protected]>

Paul Mundt

2008-07-28 10:18 UTC #5 lore raw

In reply to Yoshinori Sato

Quoted 11 lines · Yoshinori Sato

I think this problem is GCC(4.1.2) bug.
Syscall "getdents" returned "dirent->d_off" is always 0. I think other EB enviroment have same problem.
Problem code
0c03c954 <filldir>:
:
 c03c97a:       58 f7           mov.l   @(28,r15),r8 !-> offset (high)
 c03c97c:       59 f8           mov.l   @(32,r15),r9 !-> offset (low)
 c03c97e:       53 f9           mov.l   @(36,r15),r3
 c03c980:       54 fa           mov.l   @(40,r15),r4

Quoted 7 lines

:
c03c9a0: 21 82 mov.l r8,@r1 !offset(high) -> dirent->d_off
It's workaround patch.
Signed-off-by Yoshinori Sato <[email protected]>

Good catch! Applied, thanks.